Content uploaded by Jan Evler
Author content
All content in this area was uploaded by Jan Evler on Sep 28, 2021
Content may be subject to copyright.
Fourteenth USA/Europe Air Traffic Management Research and Development Seminar (ATM2021)
Flight Prioritization and Turnaround Recovery
Integrating Tactical ATFCM Slot Swapping into Resource-Constrained Ground Operations
Jan Evler, Michael Schultz, Hartmut Fricke
Technische Universit¨
at Dresden
Institute of Logistics and Aviation
Dresden, Germany
jan.evler@tu-dresden.de
Abstract—The SESAR ATM Master Plan describes the goal
to fully integrate airports into the ATM network, such that
airspace user operations are facilitated and related user costs
are reduced. The corresponding flight prioritization mechanisms
of the user-driven prioritization process are in the process of
being validated at several airports across Europe. This article
studies the benefits of the underlying concept of tactical ATFCM
slot swapping in relation to resource-constrained turnaround
management of an airline. A case study at Frankfurt airport
analyses different situations in which the airport is expected to
operate with reduced capacity, such that a local hub carrier can
prioritize its arrival (and departure) flights. Results indicate
that arrival slot swapping in constraints is very efficient as
long as fixed departure flights of the same aircraft obtain no
critical delays for the downstream network. In our case study,
such critical delays for the network occur at around 60 minutes
of departure delay, such that flights which are assigned with
higher delays require the additional flexibility of departure slot
swapping in order to achieve significant cost reductions. We
further find that optimal delay margins, which are calculated
with our approach, suffice for the confidential communication
of flight priorities, such that complex scoring and credit trading
systems might be omitted. In exchange, we propose a secondary
trading scheme, for which our model can define efficient slot
prices while considering operational constraints of an airline.
Keywords—flight prioritization; slot swapping; user-driven
prioritization process, turnaround management
I. INTRODUCTION
The Ration-by-Schedule (RBS) principle is currently ap-
plied by Air Traffic Flow and Capacity Management (AT-
FCM) during periods of high airport or airspace sector
capacity utilization (constraints) to allocate slot regulations
to flights according to the First-Planned First-Served (FPFS)
procedure. Despite continuous proposals to adapt or replace
this principle, it is still operationally accepted for its qualities
to minimize total delay in a constraint and preserve the equity
of impacted airlines. However, RBS typically cannot mini-
mize airline costs related to the allocation of ATFCM slots,
such that alternative or complementary measures are still
pursued by many research projects [1], [2].
This project has received funding from the German Federal Ministry
of Economic Affairs and Energy (BMWI) within the LUFO-V project
OPsTIMAL under grant agreement No 20X1711M. Further funding was
received from the SESAR Joint Undertaking under the European Union’s
Horizon 2020 research and innovation programme under grant agreement No
783287 (Engage KTN). The opinions expressed herein reflect the authors’
views only. Under no circumstances shall the SESAR Joint Undertaking be
responsible for any use that may be made of the information contained herein.
For instance, the SESAR ATM Master Plan 2020 [3] de-
scribes the goal to facilitate airline operations and reduce their
ATM related costs by achieving a full integration of airports
into the ATM network. In this approach, airports are focused
for their prominent position in airline network, being the
nodes where aircraft, crew and passengers connect between
subsequent flights. These interdependencies between flights
imply a high risk of disrupting the airline network, which
increases once some flights are subject to ATFCM regulations.
This might be a reason why ground operations constituted
the highest cause for primary departure delays in Europe
in 2019. In combination with reactionary (i.e., secondary)
delays, 77% of all delays in Europe have originated from
or been passed on via airline ground operations [4]. Another
7.2% of delays are attributed to air navigation services at
airports. Consequently, there is a high need to coordinate and
improve tactical procedures in case of schedule deviations,
such that ATFCM and airlines can equally benefit from the
applied recovery principles.
A. Status Quo
A potential solution to grant airlines increased operational
flexibility and reduce ATM related costs is the User-Driven
Prioritization Process (UDPP) [5]. It has been developed in
the SESAR framework and comprises several mechanisms
which are intended for the prioritization of important flights
by the airline during airport capacity constraints. Thus, air-
lines can define priority values for their flights in a constraint,
such that the assigned slots can be swapped among them
according to internal business interests (i.e., Fleet Delay Re-
ordering (FDR) and Enhanced Slot Swapping (ESS)). Further-
more, an airline can decide to suspend one of its early flights
in a constraint in order to protect the initially scheduled time
of another flight (i.e., Selective Flight Protection (SFP)). The
second mechanisms applies the Ration-by-Effort principle,
such that the equity of other airlines (especially of non-
participants) can be preserved even in case an airline wants
to protect the initially scheduled time of a flight without any
assigned slots during this period [6].
After initial validation, both mechanisms have been com-
plemented with the option to assign time windows to flight,
from which the assigned slot should not deviate. These
so-called ”margins” correspond to flight-specific delay cost
profiles, which include step costs once downstream transfer
connections or curfews are hit [6]–[8]. Recently, the second
validation step was passed at SESAR level and has confirmed
the potential of the UDPP mechanisms to reduce additional
costs incurred by airlines during a constraint. At the same
time, equity of non-users could be preserved, while airlines
with many flights (e.g., the local hub carrier during a con-
straint at Paris CDG airport) gained additional flexibility [9].
Given the limited possibilities for airlines with few flights in a
constraint to swap slots (which includes also large airlines at
spoke airports in their network), the portfolio of mechanisms
has been added by another option which allows equity transfer
between several capacity constraints in the format of flexible
credits [2], [10]. In contrast, for airlines with many flights in
a constraint, it was concluded that they might need support
in defining their individual flight priorities and margins.
The third validation step at SESAR level is already planned
and will consider arrival capacity constraints at Zurich (ZRH),
Paris (CDG) and Alicante (ALC) airports within a tactical
setting, i.e., airlines need to define their priorities with a cut-
off time of at least two hours prior to the estimated start of
a constraint. The prioritization of departure flights is still out
of scope of upcoming exercises.
B. Focus and Structure
Based on the conclusions of the recent UDPP validation
exercises, this article will study the cost benefits airlines can
obtain when tactical ATFCM slot swapping mechanisms are
incorporated into their operational procedures for resource-
constrained turnaround management. Thereby, we focus on
a slot swapping mechanism which incorporates the features
of the proposed UDPP concepts ESS and FDR with margins
on a tactical time horizon (i.e., flight priorities need to be
submitted at least two hours before the estimated arrival of the
first flight). The objectives thereby are twofold: First, we aim
at determining the recovery performance of our turnaround
recovery model during an airport constraint if tactical slot
swapping can be applied i) only to arrival flights and ii) to
all flights of an airline in this period. Second, we explore
how airlines may derive flight priorities and slot margins
from these internal assessments. The used methodology is
explained in Sec. II and applied in the context of a case
study with different scenarios as detailed in Sec. III. Results
of the case study analysis are presented in Sec. IV, whereby
Sec. V discusses potential indications for proposed and to-be-
developed slot trading mechanisms. Finally, Sec. VI draws
conclusions and presents an outlook onto future research
steps.
II. METHODOLOGY
The applied methodology introduces ATFCM capacity reg-
ulations related to an airport constraint into an airline-centric
and resource-constrained turnaround scheduling model. There
has been very little research on this agenda in the past, the
article of Santos et al. [7] being the only one (to the authors
knowledge) to incorporate airport capacity constraints in
relation to the airline delay management problem. However,
neither ATFCM slots nor links between arrival and departure
flights via the related aircraft ground operations have been
considered in there. Our article tackles these short-comings,
such that a capacitated pool of standard turnaround resources
(i.e., airport stands, personnel and ground handling equip-
ment) needs to be assigned to each aircraft, while the related
arrival and departure flights require ATFCM (runway) slots.
The latter condition may create further bottleneck situations
which may limit the flexibility of turnaround management,
given that the following conditions need to be considered:
•aircraft should be allocated to slots such that the shared
transfer connections between their arrival and departure
flights can be maintained, whereby transfer times depend
on the individual stand allocation;
•aircraft should be allocated to slots such that the limited
amount of available stands is surpassed at no time;
•aircraft should be allocated to a pair of arrival and
departure slots which guarantees a ground time larger
than the estimated turnaround time (even if it is reduced
by turnaround recovery options);
.
Fig. 1 further shows the underlying cost and time depen-
dencies for an airport constraint which affect three parallel
aircraft turnarounds and their associated arrival and departure
flights. Due to the reduced runway capacity during the con-
straint, the scheduled times of all flights have been assigned
with increasing deviations according to the RBS principle. If
the airline would maintain this initial sequence, two groups
of transfer passengers might miss their connections at the
airport, given that the arrival flights of aircraft m(highlighted
in magenta) and aircraft b(highlighted in blue) have been
assigned with arrival slots AS2and AS3. This would incur
local rebooking and compensation (”misconnex”) costs once
available transfer times fall below needed transfer times (i.e.,
when the transfer slack is consumed) as depicted by the
dashed magenta and blue step cost curves. Note here that
needed transfer times and the associated transfer slack may
change in correspondence to the stand allocation of aircraft.
Likewise, the scheduled off-block times of departure flights
may be shifted due to a regulation, such that new transfer
slack is induced. The same rationale applies to the turnaround
of all aircraft, such that scheduled ground buffers are con-
sumed by arrival slots while new buffer times are induced
by departure slots. In case the airline could prioritize flights,
an interesting case would arise if the airline decided to swap
aircraft mto arrival slot AS3. This would enable aircraft b
to maintain its initially scheduled time of arrival in AS2but
also make the turnaround of mcritical, given that mis still
assigned with the first departure slots DS1. If allowed so,
the airline may concurrently also swap departure slots among
these two aircraft to ease this dependency. However, it should
consider cost of departure delay in this process, which also
contain cost steps in relation to transfer slack at downstream
airports in the aircraft rotation.
A. Modelling Airline Slots in Airport Capacity Constraints
The capacity of an airport AQ is defined by the number
of runway movements per period P E. According to the
RBS principle commonly applied by central network man-
agement in periods when demand exceeds capacity, flights
Workshop –Results of Airline ATFM Coordination
Institute of Logistics and Aviation, Chair of Air Transport Technology and Logistics
Jan Evler // 22 June 2021
Slide 1
costs
of delay
time
𝐴𝑆1𝐴𝑆2𝐴𝑆3𝐷𝑆3
𝐷𝑆1𝐷𝑆2
Estimated Ground Time
Estimated Ground Time
Estimated Ground Time Ground Buffer
Scheduled Inbound PAX Connection Buffers
Ground Buffer
Scheduled Downstream PAX/ Crew/ Curfew Buffers
Cost of inbound delay
(local PAX misconnex)
Cost of outbound delay
(downstream operations)
SIBT EIBT SOBT EOBT
EIBT EIBT EOBT EOBT
Delay
Figure 1: Time dependencies between arrival and departure flights of the same aircraft and corresponding cost functions (which
relate to passenger transfer slack, crew and curfew buffers) if scheduled block times (SIBT/SOBT) are overrun. Starting from
estimated block times relating to each slot (EIBT/EOBT), our model estimates the ground time of each aircraft based on
constrained turnaround resources and generates a cost-minimal slot swapping solution such that all turnarounds are feasible
between assigned slots (which is initially not the case for the blue aircraft above).
which are planned first are served with the first slots (FPFS).
Thereby, Required Time of Arrival (RTA) and Calculated
Take-Off Time (CTOT) are the reference values for the place
of a flight in the queuing sequence.
For periods with reduced capacity at an airport (i.e., ”stress
periods”), this means that flights which are planned early on
receive regulations with little deviation, while delay is increas-
ing for later-planned flights. In the aftermath, delayed flights
which remain from the stress period have priority over those
which had been scheduled afterwards, such that a recovery
period is needed until the airport returns to a ”no-delay”
situation. Thus, a constraint consists of stress and recovery
periods [6], whereby the recovery period significantly depends
on the length and severity of the stress period as well as the
unassigned capacity afterwards (cf. Fig. 2).
Workshop Airline ATFM Coordination
Institute ofLogistics and Aviation, Chair of Air Transport Technology and Logistics
Jan Evler // 09 February2021
Slide 1
Stress Period Recovery Period
A1
B1
A2
C1
A3
B2
X1
C2
A4
X2
B3
A5
A6
C3
A7
B4
X3
A1
B1
A2
C1
A3
B2
X1
C2
A4
X2
B3
A5
A6
C3
A7
B4
X3
00 1 2 3 444444332100
Initial Plan
RBS (FPFS)
Delay
Hot Spot
Figure 2: Ration-by-Schedule principle in case of an air-
port/sector constraint.
B. Modelling Turnaround Management and Slot Swapping
All slots for arrival flights of an airline during an airport
constraint are defined as set AS, while all slots for departure
flights are defined as set DS. The assignment of an aircraft
afrom the set of all aircraft Ato an arrival slot s∈AS,
is done with zA
as ∈(0,1), whereas a departure slot s∈DS
is assigned with zD
as ∈(0,1). Each slot defines a calculated
take-off time CT OTsand a required time of arrival RT As
which are calculated according to the RBS principle for each
flight of the airline within the respective constraint. In the
baseline instance, each flight is fixed to its assigned slot from
the FPFS-sequence. In case this is infeasible because there is
not enough time for the turnaround between assigned arrival
and departure slots, alternative slots are available after the
last flight of the constraint. For the slot swapping exercises
introduced in Sec. III, the fixed assignment can be lifted.
The slot assignment problem is incorporated into a
turnaround recovery model which is formulated as an exten-
sion of the Resource-Constrained Project Scheduling Problem
(RCPSP). The turnaround recovery model aims at assigning
a limited set of turnaround standard and reserve resources to
the above-mentioned set of aircraft A, such that the airline
costs from a given disruption are minimised. Thereby, the
turnaround of each aircraft a∈Ais defined by scheduled
start and finishing times SI BTaand SOBTa, which are
adopted from the flight plan. It consists of a network of
sub-processes P, where each sub-process i∈Pis char-
acterized by the related aircraft (RAi=a), has a variable
starting time siand a duration Diwhich corresponds to the
80%-quantile of a statistically-fitted time distribution [11].
Links between turnaround sub-processes are determined in
the precedence matrix P Mij ⊆P×P. Following
the general RCPSP, each aircraft aneeds to be assigned
to an airport stand pfrom the set ST with χap ∈(0,1),
whereby we differentiate between contact stands CS and
remote stands RS, which are all equipped with the nec-
essary personnel and resources for a standard turnaround.
Thus, the in-block process i∈IB ⊂P, as the first
process of each turnaround, can only be scheduled if a
stand is available which fulfills all operational requirements
for aircraft and flights. It further depends on the estimated
landing time eldtain the assigned slot and the average taxi-
in duration EX I T . Downstream sub-processes can only start
once all preceding processes are (scheduled to be) completed.
Thereby, turnaround reserve resources enable schedule recov-
ery options, such as a quick turnaround (ωi∈(0,1)), which
reduces the duration of specific turnaround sub-processes,
e.g., cabin cleaning i∈CL ⊂P. These reserve resources
are limited by QT R and incur recovery costs Cqta
iby each
application. Another recovery options is stand reallocation
which considers that aircraft which are positioned at a remote
stands p∈RS have reduced de-boarding and boarding
duration, given that passengers can use front and rear doors.
It further considers that the stand location of arrival and
departure aircraft at an airport directly influences the needed
transfer time N T Tij for connecting passengers. By applying
these options, airlines can influence the total duration along
the critical path of a turnaround but also time dependencies
between aircraft along transfer processes i∈P A ⊂P. If
a transfer process would require a departure flight to delay
its off-block, the airline can either cancel the connection
(κi∈(0,1)) or accept the delay. The prior decision would
incur costs of care, rebooking and compensation Ccnx
i. In
case the transfer involved a crew, it can only be cancelled
if a standby is available. Delaying the departure of a flight
results in marginal linear costs for crew wages, maintenance
and passenger dissatisfaction Clin
al . Furthermore, departure
delay might disrupt transfer processes at downstream airports
in the aircraft rotation, which incur step costs Cstp
al once
the slack before this critical event is consumed (determined
by yas ∈(0,1)). The estimated departure delay is captured
within delay levels l∈Lafter the scheduled off-block
time SOBTa. The resulting estimated off-block time eobtais
added with an average taxi-out duration EXOT to calculate
the estimated take-off time etota, which needs to comply with
the CT O Tsof the slot sassigned to aircraft a.
C. Mathematical Formulation
min X
a∈A
X
l∈L
Clin
al ral +Cstp
al yal +
+X
i∈P
Cqta
iωi+Ccnx
iκi(1)
s.t. si≥eldta+EX IT ∀i∈I B |RAi=a(2)
eldta≥RT As−5 + M(1 −zA
as)∀a∈A;∀s∈AS (3)
eldta≤RT As+ 10 + M(1 −zA
as)∀a∈A;∀s∈AS (4)
X
s∈AS
zA
as = 1 ∀a∈A(5)
X
a∈A
zA
as ≤1∀s∈AS (6)
si≤SOB Ta+X
l∈L
ral =eobta∀i∈OB |RAi=a(7)
eobta+EX OT =etota∀a∈A(8)
ral ≥(UBal −LBal )yal ∀a∈A;∀l∈L(9)
ral ≤(UBal −LBal )ya(l−1) ∀a∈A;∀l∈L(10)
etota≥CT OTs−5 + M(1 −zD
as)∀a∈A;∀s∈DS (11)
etota≤CT OTs+ 10 + M(1 −zD
as)∀a∈A;∀s∈DS (12)
X
s∈DS
zD
as = 1 ∀a∈A(13)
X
a∈A
zD
as ≤1∀s∈DS (14)
sj≥si+Di∀i∈IB , ∀j∈P|PMi,j = 1
(15)
sj≥si+Di(1 −χap) + αDiχap
∀i∈DE ∪BO , ∀j∈P|
P Mi,j = 1; RAi=a;∀p∈RS
(16)
sj≥si+NT Tij χap χbq −M κi
∀i∈P A, ∀j∈P|P Mi,j = 1,
RAi=a, RAj=b, ∀p, q ∈ST
(17)
X
p∈ST
χap = 1 ∀a∈A(18)
X
a∈A∪A0
xSabp =χap ∀b∈A;∀p∈ST (19)
X
b∈A∪A0
xSabp =χap ∀a∈A;∀p∈ST (20)
sj≥si+T−M(1 −xSabp)
∀a∈A;∀b∈A∪A0; ∀p∈ST ;
∀i∈OB |RAi=a;
∀j∈IB |RAj=b
(21)
sj≥si+Di(1 −ωi) + βDiωi∀i∈C L, ∀j∈P|PMi,j = 1
(22)
X
b∈A∪A0
xQab ≤QT R ∀a∈A0(23)
X
a∈A∪A0
xQab =ωb∀b∈A(24)
X
b∈A∪A0
xQab =ωa∀a∈A(25)
si≥eobta+T−M(1 −xQab)∀a∈A;∀b∈A∪A0;
∀i∈IB |RAi=b
(26)
The objective function (1) minimizes total costs of delay
and schedule recovery. This includes linear costs across all
delay levels, step costs once a critical delay thresholds is
overrun and costs related to turnaround recovery and cancel-
lation of transfer connections. The start of each turnaround
can only scheduled after landing and taxi-in of the arrival
flight (2), whereby the estimated landing time must align with
the assigned arrival slot (3)-(4). Arrival slots typically consists
of an required time of arrival with a grace period of minus
five and plus ten minutes. All arrival flights which are part
of an airport constraint need to be assigned to exactly one
arrival slot (5), while each slot can be used by maximum one
flight (6). If the estimated turnaround off-block time exceeds
the SOBT, departure delay is distributed across pre-defined
delay levels (7) and is translated into an estimated take-off
time (8). Each delay level is bounded such that delay can only
occupy upper levels by taking into account the related step
costs before them (9)-(10). Furthermore, constraints (11)-(14)
consider that an aircraft can only be released ”off-block”, if
a departure slot is available.
Standard scheduling constraints (15) ensure that all
turnaround sub-processes following on the in-block process
can only start once it has been finished. Similar scheduling
constraints are defined for processes starting after deboarding
and boarding (16), whereby the duration of both process can
be reduced by factor αwhen an aircraft is positioned at
a remote stand p∈RS. Constraints (17) consider needed
transfer times for connecting passengers and crews between
the stands of their arrival and departure flights, which di-
rectly influences their stand allocation. Hereby, note that the
quadratic formulation needs to be linearized for the applica-
tion of standard solvers (as described in [12]). Further note
that this dependency is omitted for all transfer connections
which are cancelled. Following the RCPSP, constraints (18)
makes sure that each aircraft is allocated to exactly one stand,
whereby the MTZ-formulation in constraints (19)-(21) defines
the sequence of aircraft which use equal stands whereby
dummy node A0marks the start and end of each sequence.
The standard RCPSP formulation is extended by the pos-
sibility to assign turnaround recovery resources to some sub-
processes (e.g., cabin cleaning) which then reduce the respec-
tive durations by factor β(22). Considering that turnaround
recovery resources are limited (23), another MTZ-formulation
in constraints (24)-(26) builds a sequence which ensures that
only so many turnarounds can be prioritized in parallel as
recovery resources are available.
III. SCE NAR IO A ND AP PL IC ATIO N
The integrated turnaround recovery and slot swapping
model from Sec. II is applied in the context of an airline case
study network with hub at Frankfurt airport. This sections
presents the case study setting and introduces scenarios which
include a airport capacity reduction during the morning hub
bank. For each scenario, a new FPFS-sequence is calculated
such that the applied turnaround recovery must adhere to the
assigned arrival and departure slots. In further course, fixed
assignments for arrival slots or respectively for all slots are
lifted, such that we can assess the recovery performance of
the model with tactical slot swapping.
A. Case Study Setting
A flight schedule was adopted from the summer sched-
ule 2019 of a local hub carrier and comprises 15 parallel
turnarounds during the morning hub bank (i.e., 7:30 a.m.
to 11:00 a.m. local time). Between the related 30 flights
(cf. Fig. 3), passenger connections are simulated to resemble
potential itineraries which adhere to the average connection
ratio (55% transfer passengers) and minimum connecting
time in Frankfurt (45 minutes), a typical seat load factor
of a hub carrier (85%) and avoid extreme detours (e.g.,
passengers from Madrid are unlikely to connect via Frankfurt
to Barcelona). Crew connections and the initial stand allo-
cation are generated with separate optimisation algorithms,
such that they comply with official operational constraints
and minimize crew assignment costs and the total needed
transfer time respectively. Contact stands in Terminal 1A (cf.
Stands A1, A2, A3 and A5 in Fig. 4) are reserved for flights
to and from Schengen countries only. Contact stands with
special security and customs areas (cf. Stands A3, A6, B1
and C1) can also operate flights to and from non-Schengen
countries. Stands A3 and A6 are predominantly used for
intercontinental flights with wide-body aircraft. Stands R1
and R2 are remote, which means that passengers need to
be transferred with apron buses via the central bus stations
(marked with a bus icon in Fig. 4).
CAI
BEG
MAD
KRK
VCE
PMI
ARN
FCO NAP
CDG
HAM
TXL
MUC
SEA
EWR
ORD
SEA
IST
DUB
TLV
ZRH
RIX
LED
BCN
LHR
LYS
SCQ
Figure 3: Initial Flight Plans (M1) as submitted by the case
study airline for arrival (red) and departure flights (blue) to
and from Frankfurt airport.
Master UseCase –OPsTIMAL Projekt
Institut für Luftfahrt und Logistik, Professur Technologie und Logistik des Luftverkehrs
Prof. Dr.-Ing. Hartmut Fricke, Dr.-Ing. Henning Preis, Jan EvlerM.A.
N
Frankfurt Main Airport FRA
Schengen Stand
Non-Schengen Stand
07:30 08:00 08:30 09:00 09:30 10:00 10:30 11:00
A1
A2
A3
A4
A5
A6
B1
C1
R1
R2
Airport Stands
Initial Stand Allocation
PMI (8) A320 ZRH
VCE (3) A320 LYS FCO (10) A321 DUB
MUC (15) A320TXL
MAD (9) A321 BCN TXL (14) A320 SCQ
CDG (11) A320 HAM
EWR (1) B748 ORD
SEA (7) B744 SEA
KRK (2) A320 IST NAP (12) A320 RIX
ARN (6) A320LED
BEG (4) A320 LHR
CAI (5) A320 TLV
HAM (13) A320ARN
A1
A2
A3
A4
A6
A5
B1
C1
R1
R2
Figure 4: Case study setting at Frankfurt airport (FRA).
Three aircraft, including two wide-bodies, are fixed at
their initial stands due to operational constraints, whereas the
remaining twelve aircraft can be re-allocated to any other
stand complying with the required security procedures of the
respective arrival and departure flights. In total, one quick
turnaround unit (QT R = 1,Cqta
i= 500 per turnaround)
and one standby crew (Ccnx
i= 1000; two additional wage
hours) enable the respective schedule recovery options. The
cost of cancelling a passenger transfer Ccnx
iare adopted
from reference values per passenger as determined in [13]
and consider the additional trip time on the next alternative
flight as well as that some passengers may not wish to be
re-booked and need to be compensated and/ or reimbursed
instead according to EU regulation 261. Note that we consider
compensation by the airline as a worst case scenario, although
the airline is not liable for ATFCM delays - but might be
if it purposely changes the flight sequence and the assigned
delays. Marginal costs of departure delay Clin
al are constant
per delay level and are also adopted from reference values
as determined in [14]. They include additional crew wages,
maintenance expenses and costs of passenger dissatisfaction
in relation to the respective aircraft type. Step costs related
to departure delay Cstp
al are calculated in the same way as
local costs of rebooking and consider the downstream flight
schedule of each aircraft and the related transfer processes at
other airports with their respective slack time. Together they
result in flight-specific delay cost functions (cf. Fig. 8).
B. Scenarios
The runway capacity at Frankfurt airport is defined with
108 movements per hour, which corresponds to an standard
capacity of AQ = 27 per period P E = 15 minutes.
The planned capacity utilization is marked by the green
line in Fig. 5 and was retrieved from the initial flight plan
data (M1) on Friday, 16 August 2019 - a day which did not
contain any capacity regulations at Frankfurt airport. Based
on this initial flight sequence, three constraint scenarios are
introduced, which each predict the runway capacity to reduce
to AQ = 15 while the length of the stress period varies.
All scenarios assume that all flights will be operated, such
that potential cancellations are neglected. The stress period is
estimated to start before the morning peak at 7:00 a.m. local
time. In Scenario 1 (S1), it is estimated to last for two
hours, while in Scenario 2 (S2) it includes three hours and
in Scenario 3 (S3) four hours. As demand exceeds capacity
during the entire stress period, slots are assigned according
to the RBS principle. This results into a recovery period
until 2:00 p.m. in S1 which also covers the entire midday
bank. Given that more flights are affected by the stress
period in S2, the recovery period covers almost the entire
afternoon bank until 5:30 p.m.. In S3, it lasts for another
75 minutes (cf. Fig. 5). Based on this initial slot assignment,
the calculated landing and take-off times per scenario are
retrieved per flight of the case study airline and define sets of
arrival slots AS and departure slots DS. Considering a cut-off
time two hours prior to the start of the constraint, we assume
a look-ahead time of three hours before the problem, such that
there is one hour left for the airline to define the priorities
of their flights within their assigned slots. Each scenario is
run with two instances, whereby the first one (S1/2/3-A)
allows the airline to swap only arrival flights within their
slots, whereas in the second one (S1/2/3-A+D), arrival and
departure flights can be swapped within their respective sets.
Note that three hours before the constraint, three flights from
Cairo (CAI), Newark (EWR) and Seattle (SEA) have already
departed their airports of origin. Thus, they are fixed within
their assigned arrival slots.
Workshop Airline ATFM Coordination
Institute ofLogistics and Aviation, Chair of Air Transport Technology and Logistics
Jan Evler // 09 February2021
Slide 1
Stress Recovery
Stress Recovery
Stress Recovery
Scenario 1:
Scenario 2:
Scenario 3:
Figure 5: Airport constraint scenarios during the morning hub
bank at Frankfurt airport.
Further note that from an airline perspective, runway slots
in a capacity constraint need to be considered in pairs for
arrival and departure flights of the same aircraft. This is due
to the fact that both flights might be assigned with different
delays, which is especially critical when the departure slot
receives less delay and the scheduled turnaround time is very
tight. In this case, it might be infeasible for the departure
flight to use its assigned slot given that the higher arrival
delay cannot be absorbed during the turnaround. Thus, a
new departure slot needs to be requested, which might be
far down in the sequence, especially when considering the
long airport recovery period. The phenomenon of diverging
delays is already visible in the exemplary sequence exhibited
in Fig. 2, such that delays increase on flights early in the
stress period, stagnate at the end of the stress period and
begin to decrease once free capacity is available during the
recovery period. A similar pattern can be found in all three
scenario instances (cf. Fig. 6), where the effect is enforced
by very heterogeneous ground times, such that the arrival
sequence of aircraft significantly deviates from the departure
one. In S1, almost all arrival flights receive higher delays
than the departure flights of the respective aircraft. Thereby,
the flights of aircraft 7 and 11 obtain the largest difference
(i.e., 25 minutes), which is critical for aircraft 7, due to no
scheduled ground buffer, but uncritical for aircraft 11, which
has 75 minutes ground buffer. In S2, aircraft 11 to 15 have
diverging arrival and departure delays, whereas in S3, only
aircraft 14 and 15 are concerned. Thereby, aircraft 12 to 15
have very few to no absorption potential for arrival delays,
given that their turnaround is scheduled for 45 and 50 minutes.
Consequently, both departure flights require new departure
slots, which are assumed to be available at 12:00 p.m. and
12:30 p.m..
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Aircraft
0
10
20
30
40
50
60
70
80
90
Slot Deviation from Schedule in Minutes
S1 Inbound
S2 Inbound
S3 Inbound
S1 Outbound
S2 Outbound
S3 Outbound
Figure 6: Slot deviation from schedule per aircraft and sce-
nario according to the RBS principle.
IV. ANA LYSI S
The turnaround recovery and slot swapping model was
solved with IBM CPLEX Version 12.10.0-0 on a 4-core CPU
with 8GB RAM. Aligning to the tactical decision horizon,
the solution process was aborted after one hour if optimality
of a solution could not be proven and there was still a gap
between upper and lower bounds. A gap remained for some
instances from the reference case (i.e., no turnaround recovery
and no slot swapping) and the full slot swapping model (cf.
Tab. I). This indicates that the turnaround recovery model
with integrated slot allocation (arrival + departure) and stand
allocation is np-hard, given that each of the three individ-
ual allocation problems can be related to the Generalized
Assignment Problem (GAP), which itself is np-hard [15].
This may explain why even in the reference case, where
the stand allocation is fixed, a gap remains within reasonable
solution times. Due to the nature of the problem, there are
too many partially symmetric options with similar objective
values, which form a weak lower bound, such that standard
solvers need to test all feasible options before converging to
an optimal solution. In operational practice, this induces a
managerial trade-off whether a solution should be optimal
- may require very long solution time or development of a
sophisticated solution method - or optimized - not proven to
be optimal but close-to-optimal (relation to gap) - and anyway
better than without the optimization process.
A. Total Airline Costs
Tab. I further exhibits total costs incurred by the airline
per airport constraint scenario. Total costs rise exponentially
in the reference cases from S1 to S2 and S3, which relates
to higher assigned delays in the FPFS-sequence of S2/S3 that
do not consider flight-specific cost drivers. The turnaround
recovery model with integrated arrival slot swapping is very
effective for shorter capacity constraints, such that total costs
can be reduced by 85% in S1 and 57% in S2. In both
scenarios, increased flexibility from departure slot swapping
would result in only marginal additional costs savings of
1-2%. Given that in both cases the assigned delays range
between 10 and 60 minutes, there is sufficient absorptive
capacity to compensate delays during ground operations -
especially when considering that less than 30% of all pas-
senger connections have small slack of 30 minutes or less,
while the average ground buffer is 25 minutes. In combination
with the capacity of the recovery model to adapt the stand
and arrival slot allocation, most of the critical connections
can be saved. Furthermore, it needs to be taken into account
that cost of departure delay are comparatively small on most
flights during the first 60 minutes, such that a reallocation of
the initially assigned departure slots in S1-A+D and S2-A+D
does not result in high cost saving effects (cf. Fig. 8).
Conversely in S3, many of the assigned departure delays
according to the RBS principle surpass cost-intensive thresh-
olds which relate to disruptions in the downstream network
of the case study (cf. Fig. 8). Consequently, swapping arrival
slots while departure slots remain fixed, brings only limited
cost benefits for the airline (i.e., 11% - cf. Tab. I). If this
fixation would be lifted for departure flights, total costs in S3-
A+D can be reduced by at least 41%.
TABLE I. Total airline cost per scenario including required
solution time/ remaining gap after 3600s.
ID Reference FPFS Arrival SSW Arrival+Dep. SSW
S1 54899 (1.17%) 8131 (260s) 7568 (2710s)
S2 115718 (0.61%) 49840 (0.40%) 47511 (2448s)
S3 205463 (15s) 183725 (326s) 121764 (16.03%)
B. Slot Allocation
As mentioned in Sec. III-B, it may be critical for airlines
if a departure flight receives less delay during a capacity
constraint than the arrival flight of the same aircraft. Once
this is the case, airline have three options: 1) return the
assigned (departure) slot and request a new one; 2) accelerate
the turnaround; and/or 3) swap the slots such that there is
sufficient ground time between both flights. Given that the
latter two options are not available in the reference case, three
departure slots cannot be used by the airline in S1, while in S2
and S3 the airline needs to request one new slot respectively.
Conversely, in all instances that involve slot swapping, the
airline can operate with all initially assigned slots and no
additional slots need to be requested.
Fig. 7 highlights the optimized arrival slot allocation per
scenario instance. When only arrival slot swapping is allowed,
the optimal sequence remains stable across all three scenarios,
although there being some alternative optimal solutions with
different sequences. This indicates that there is a robust
optimal arrival sequence for the airline in our case study,
despite the fact that later slots obtain higher delays in S2 and
S3. However, in case that arrival and departure slots can be
swapped, the optimal arrival sequence changes significantly
within each scenario and between scenarios. Based on that,
we derive a strong correlation between the assigned arrival
and departure flight slots of an aircraft, such that the robust ar-
rival sequence in the first instances relates to the fixed FPFS-
sequence for all departure flights. Conversely, the additional
flexibility induced by departure slot swapping requires the
calculation of an individual sequence per constraint, which
can hardly be generalised if the delays within a constraint
are changing (e.g., due to uncertainty about the length of the
stress period). Note that none of the flights remains within its
initially assigned FPFS-slot throughout all instances (except
for all flights which already departed and are fixed within their
slots). Some aircraft are prioritized in most instances (e.g.,
aircraft 3 [orange], aircraft 6 [maroon], aircraft 9 [cyan]),
while others are typically suspended (e.g., aircraft 2 [ochre],
aircraft 11 [light blue]).
7:25 8:05
08:25
08:35
08:55
09:05
09:15
09:15
09:20
09:30
09:35
09:45
10:00
10:20
10:25
7:25 8:05
08:25
08:35
08:55
09:05
09:15
09:20
09:25
09:45
09:50
10:05
10:25
10:45
10:50
7:25 8:05
08:25
08:35
08:55
09:05
09:15
09:20
09:25
09:45
09:50
10:05
10:35
11:10
11:15
FPFS (1-15)
Scenario 1
Arrival Slot
Swapping
Arrival +
Departure
Scenario 2
Arrival Slot
Swapping
Arrival +
Departure
Scenario 3
Arrival Slot
Swapping
Arrival +
Departure
Figure 7: Optimized sequence of arrival flights within as-
signed airline slots per scenario.
For reasons already stressed in the previous section, there
are only few departure slot swaps in S1-A+D and S2-A+D
(i.e., 3 and 4). However in S3-A+D, almost all departure
flights are swapped among each other such that critical
cost steps are avoided. Thereby, three departure flights are
suspended to later slots (e.g., aircraft 2 from the second
position to the last in the sequence) while ten flights receive
185
Figure 8: Assigned departure delay per flight in the FPFS-sequence and in the optimized recovery solution in comparison to
departure delay cost functions.
earlier slots than assigned in the FPFS-sequence (cf. Fig. 8).
This model output corresponds largely to the principle of the
SFP mechanism - without using a credit system - such that the
suspension of one early flight can mitigate delays on many
others while the overall delay remains equal.
C. Applied Turnaround Recovery Options
Tab. II lists the optimal number of applied turnaround
recovery options and derives the relative change in average
passenger delay per scenario. Note that in S1, a quick-
turnaround procedure is assigned to aircraft 7, given that
the arrival slot obtains 25 minutes more delay than the
departure slot and the scheduled ground time comprises no
buffer (cf. Fig. 6). Likewise, in S2-A and S3-A, a quick-
turnaround is required for aircraft 14 which obtains more
arrival than departure delay in the FPFS-sequence. However,
once departure slot swapping is allowed, the optimal solution
in S2-A+D includes a quick-turnaround for aircraft 11 (which
has a large ground buffer but is suspended in the arrival
sequence), while in S3-A+D, the operation of the turnaround
with additional personnel is not efficient for any aircraft.
Stand changes are required throughout all instances and help
to avoid missed passenger connections when only arrival slot
swapping is allowed. Thereby, the optimal stand allocation
relates to the optimal arrival sequence, such that no generally
optimal scheme can be found for all scenarios. In case of
combined arrival and departure slot swapping, up to 13 out
of 151 passenger transfers need to be re-booked in order to
allow a more comprehensive reordering of the initial flight
sequences. Given that delay increases significantly for re-
booked passengers (depending on the departure time of the
next available flight towards the same destination), average
delay increases by up to 2.6%. Though, average delay for
passengers on their original flights decreases by up to 3.6%.
TABLE II. Optimal number of applied turnaround recovery
options per scenario.
ID Quick Turnaround # Stand Changes # Misconnex Delay p.PAX
S1-A Aircraft 7 8 0 -0.4%
S1-A+D Aircraft 7 9 9 +1.7%
S2-A Aircraft 14 7 0 +0.3%
S2-A+D Aircraft 11 4 6 +2.6%
S3-A Aircraft 14 8 0 +0.1%
S3-A+D - 8 13 +0.6%
D. Definition of Flight Priority Values and Margins
From the perspective of an airline, we see three ways
to determine priority values for flights in a constraint: 1) a
qualitative approach which relates the ranks of flights in the
optimal sequence to each other; 2) a quantitative approach
which weights the assigned optimal delay of flights among
each other; and 3) a heuristic approach which compares the
flexibility of all applicable flights for a slot. For the first ap-
proach, we apply a version of the Analytic Hierarchy Process
(AHP) to the optimal flight sequences [16]. Therefore, the
optimal ranks of all flights from the arrival and departure slot
swapping instances (S1/2/3A+D) are rated pair-wise against
each other, such that, e.g., a flight which obtained the first
rank in the arrival or departure sequence dominates another
flight which was assigned with the fifth rank by factor 4, while
the latter flight receives the inverted factor 1/4. Despite of
the different scoring system from 1 to 14 (the original AHP
assigns priority values in the range between 1 and 9 between
two attributes), the resulting matrix shows a low inconsistency
value of less than 7%, which is acceptable [16]. The resulting
rankings per scenarios are than multiplied with another by
assuming equal probabilities for each of the three scenarios
to derive a general sequence also for instances in which
all slots can be swapped. The generalized sequence is then
implemented with fixed slot assignments into the respective
modelling instances and results in ”optimal” total costs which
are 37% higher than in S1-A+D (27% in comparison to S1-A)
and 7% higher than in S2-A+D (2% in comparison to S2-A)
while in S3 it is infeasible due to conflicts with the limited
stand capacity. Thus, as already concluded in Sec. IV-B, there
is no generally optimal (robust) flight sequence if arrival and
departure slots can be swapped among each other.
For the second approach, we assign priority values to flights
based on the delay they obtain with their optimal slot. Thus,
if a flight is assigned to a slot close to its initially schedule
arrival or departure time, it has a high priority score, while
priority values are decreasing with higher delays. Thereby, we
consider equivalent delays between all priority scores which
are determined by the difference of the highest and the lowest
assigned delay within a set, divided by N−1(Nbeing the
number different priority values).
Fig. 9 displays the resulting priority values for arrival
flights in scenario S1-A. Similar to the UDPP, 1 marks the
highest flight priority, 9 the lowest, three flights are fixed (f),
while each flight cannot be assigned to any slot before its
scheduled time of arrival. Note that only few mid-range
values appear in this instance, which was also found in UDPP
validation exercises [17]. Further note that assigning slots to
flights according to these priority scores - by respecting FPFS
for flights with equal scores - does not yield the optimal flight
sequence if the delay is unbounded for all flights (cf. Fig. 7).
Workshop Airline ATFM Coordination
Institute ofLogistics and Aviation, Chair of Air Transport Technology and Logistics
Jan Evler // 09 February2021
Slide 1
Aircraft
07:25
07:30
07:35
07:40
07:45
07:50
07:55
08:00
08:05
08:10
08:15
08:20
08:25
08:30
08:35
08:40
08:45
08:50
08:55
09:00
09:05
09:10
09:15
09:20
09:25
09:30
09:35
09:40
09:45
09:50
09:55
10:00
10:05
10:10
10:15
10:20
10:25
1 f
2 88888888888888888888888888888888
3 111111111111111111111111111111
4 99999999999999999999999999999
5 f
6111111111111111111111111
7 f
88888888888888888888888
9 8888888888888888888888
10 777777777777777777
11 888888888888888888
12 777777777777777
13 777777777777
14 7777777
15 5 5 5 5 5 5
1 3 2 6 5 10 7 2 13 8 9 15 14 4
12
Figure 9: Flight priorities without margins in S1-A.
However, once the corresponding delay from the assigned
optimal slots is incorporated as upper bound of the respective
flight margin, the optimal sequence can be derived using the
following algorithm:
1) Assign flights with fixed slots;
2) Assign flights with from hightest to lowest priority
scores (* if several with same score, first those with
least margin after slot, then FPFS);
a) if no slot is available, break assignment in latest
possible slot for this flight and assign there;
b) assign detached flight to next possible slot;
For the third approach, we consider that priority scores and
margins are both based upon the optimal amount of delay,
such that priority scores may also be omitted if optimal
delay margins are used. Each slot is then assigned with a
simple heuristic to the applicable flight with the least available
margin (cf. Fig. 10). The heuristic without priority scores
but margins only was tested for all arrival and departure
sequences and always yielded the optimal sequence.
Workshop Airline ATFM Coordination
Institute ofLogistics and Aviation, Chair of Air Transport Technology and Logistics
Jan Evler // 09 February2021
Slide 1
Aircraft
07:25
07:30
07:35
07:40
07:45
07:50
07:55
08:00
08:05
08:10
08:15
08:20
08:25
08:30
08:35
08:40
08:45
08:50
08:55
09:00
09:05
09:10
09:15
09:20
09:25
09:30
09:35
09:40
09:45
09:50
09:55
10:00
10:05
10:10
10:15
10:20
10:25
1f
2
3
4
5f
6
7f
8
9
10
11
12
13
14
15
1 3 2 6 5 4 7 8 10 11 12 13 15 14
9
Figure 10: Flight priorities with margins in S1-A.
V. DISCUSSION ON CREDITS AND SL OT TRADING
The currently incorporated slot swapping mechanism grants
flexibility to airlines without compromising the equity of other
users, given that the overall delay remains equal. However,
its efficiency may depend on a certain market share within
an airport capacity constraint, as otherwise the available
swapping options are very limited. The UDPP framework
enhanced the SFP mechanism for this issue, such that low-
volume users were envisioned to trade slots into credits and
transfer them into other constraints. Such complex regula-
tions tend to contradict initial intentions and may create
new problems as a result of the developed solutions, e.g.,
credit systems are prone to human behavioural biases [18].
Furthermore, simplified approaches tend to be more stable
during operations (i.e., controller can follow decisions). Thus,
similar to omitting flight priority values, we aim at creating a
simple trading mechanism, which is based on the secondary
trading scheme instead of credit values. Thereby, we consider
that sometimes airlines may prefer to maintain a transfer
connection instead of using their assigned slots. Consequently,
we conducted an small experiment in which we neglect
an assigned departure slot at 9:00 a.m. in the optimization
process, such that we analyse the additional costs the airline
would incur by using an alternative slot at the end of the
sequence instead. These additional costs may then represent
the minimum price another user needs to pay for trading
the released slot. Conversely, we consider also an additional
departure slot at 8:10 a.m. (released by another airline) in our
model in order to determine the associated extra cost savings,
i.e., the maximum price our airline should pay for using this
slot. Tab. III lists the related changes in total costs, which are
volatile per scenario instance and depend on the position of
the slot in relation to all other slots, operational restrictions
of turnaround resources and the amount of delay that can be
saved by a bought slot/ is added by an alternative slot at the
end of the sequence.
TABLE III. Change of total costs in case of secondary trading.
ID Sold Dep Slot ∆Total Costs Bought Dep Slot ∆Total Costs
S1-A 9:00 +12448 8:10 0
S1-A+D 9:00 +13046 8:10 -250
S2-A 9:00 +58031 8:10 -4842
S2-A+D 9:00 +15268 8:10 -4829
S3-A 9:00 +17715 8:10 -53855
S3-A+D 9:00 +12343 8:10 -22676
VI. CONCLUSION
In this paper we incorporated a tactical ATFCM slot
swapping mechanism into a resource-constrained turnaround
scheduling model, such that airlines can calculate an opti-
mized flight prioritization and resource allocation strategy
during airport capacity constraints. Within a case study of
a hub airline at Frankfurt airport, we find that swapping only
arrival flights within their assigned slots is very efficient as
long as the assigned delays do not exceed critical cost thresh-
olds which correspond to downstream network disruptions.
In the analysed airline case study network these thresholds
appeared at around 60 minutes, but might differ in other
airline networks. Leaving departure slots fixed as defined
in the FPFS-sequence, yields a stable optimal arrival flight
sequence, which seems to be unaffected by the length of a
constraint and the related volatile delay in some slots.
Departure slot swapping and the corresponding additional
flexibility for airlines brings only marginal cost benefits for
constraints with small and medium delay below the critical
threshold. These benefits are likely to be compensated by
administrative efforts, given that due to an increased solution
space no robust optimal sequence can be calculated across
different constraints. Thus, the parallel prioritization of arrival
and departure flights would require an airline to make case
specific calculations, which may only be efficient in high
delay scenarios, when delays exceed the critical threshold
and cause many cost-intensive downstream disruptions. Note
that in these instances also the complexity of the problem
increases, which is np-hard, such that an optimal solution may
not always be found within reasonable time. Future research
should increase the number of studied constraint scenarios to
test if these findings can be generalized, e.g., by determining
the efficiency of slot swapping for airlines as a function of
the magnitude of assigned delays.
In any case, our simple slot swapping mechanism has
demonstrated the capacity to grant airlines additional flexibil-
ity and significant cost savings during airport constraints. This
comes without any complexities such as credit systems or
priority scores as initially envisioned by the UDPP. Priorities
in-between flights can be defined by assigning maximum
delay margins to each flight, which ensures a confidential
way of communication between ATM stakeholders. Unused
slots can be traded directly with other airlines, whereby our
models can calculate a minimum price for selling a slot
and a maximum price for buying a slot. The opportunity to
define such efficiency bounds for secondary slot trading or
auction systems may provide incentives for an early release
of unused slots. Future research should study how this may
benefit airlines with a low volume of flights in a constraint.
As a necessary pre-condition to our approach, airlines need
to integrate ground operations into their schedule recovery
procedures [12] and define flight-specific delay cost functions
in correspondence to the scheduled downstream operations of
each aircraft [8]. Further note that not all schedule recovery
options are considered in our current model, given that aircraft
swaps and flight cancellations are neglected. Future research
may include these options and study our approach also outside
of airport constraints such that tactical slot swapping may
contribute to daily airline recovery procedures.
ACKNOWLEDGMENT
The authors would like to thank Eurocontrol for the pro-
vision of initial flight plan data which have been used as
baseline for the flight sequence at Frankfurt airport on the
scenario day in August 2019.
REFERENCES
[1] M. Ball, C. Barnhart, G. Nemhauser, and A. Odoni, “Air Transporta-
tion: Irregular Operations and Control,” in Handbooks in Operations
Research and Management Science, 2007, vol. 14, pp. 1–67.
[2] S. Ruiz, L. Guichard, N. Pilon, and K. Delcourte, “A New Air Traffic
Flow Management User-Driven Prioritisation Process for Low Volume
Operator in Constraint: Simulations and Results,” Journal of Advanced
Transportation, vol. 2019, pp. 1–21, Apr. 2019.
[3] Eurocontrol, “European ATM Master Plan 2020.pdf,” SESAR Joint
Undertaking, Luxembourg, Tech. Rep., 2019.
[4] ——, “CODA Digest 2019,” Eurocontrol CODA, Tech. Rep., 2020.
[5] N. Pilon, K.-D. Hermann, A. Ds, P. Hlousek, M. Hripane, E. P. Parla,
F. Rousseau, D. Chiesa, G. Q. Albert, and D. Muller, “PJ07 Final
Project Report,” Eurocontrol, Tech. Rep., 2019.
[6] N. Pilon, A. Cook, S. Ruiz, A. Bujor, and L. Castelli, “Improved
Flexibility and Equity for Airspace Users During Demand-capacity
Imbalance,” in 6th SESAR Innovation Days, Delft, 2016, p. 8.
[7] B. F. Santos, M. M. Wormer, T. A. Achola, and R. Curran, “Airline de-
lay management problem with airport capacity constraints and priority
decisions,” Journal of Air Transport Management, vol. 63, pp. 34–44,
Aug. 2017.
[8] J. Evler, M. Schultz, H. Fricke, and A. Cook, “Development of
Stochastic Delay Cost Functions,” in 10th SESAR Innovation Days,
online, 2020, p. 9.
[9] N. Pilon, L. Guichard, and K. Cliff, “Reducing Impact of Delays using
Airspace User- Driven Flight Prioritisation,” in 9th SESAR Innovation
Days, Athens, 2019, p. 8.
[10] S. Ruiz, L. Guichard, and N. Pilon, “Optimal Delay Allocation under
High Flexibility Conditions during Demand-Capacity Imbalance,” in
7th SESAR Innovation Days 2017, Belgrade, 2017, p. 28.
[11] H. Fricke and M. Schultz, “Delay Impacts onto Turnaround Perfor-
mance,” in 8th USA/Europe Air Traffic Management Research and
Development Seminar (ATM2009), 2009, p. 10.
[12] J. Evler, E. Asadi, H. Preis, and H. Fricke, “Airline Ground Oper-
ations: Schedule Recovery Optimization Approach with Constrained
Resources,” Transportation Research Part C: Emerging Technologies,
p. 31, 2021.
[13] A. Cook and G. Tanner, “The cost of passenger delay to airlines in
Europe,” University of Westminster, London, Consultantion Document,
2015.
[14] A. Cook, “European airline delay cost reference values - updated and
extended values,” University of Westminster, London, Tech. Rep. 4.1,
2015.
[15] M. L. Fisher, R. Jaikumar, and L. N. Van Wassenhove, “A Multiplier
Adjustment Method for the Generalized Assignment Problem,” Man-
agement Science, vol. 32, no. 9, pp. 1095–1103, Sep. 1986.
[16] T. L. Saaty, “Decision making with the analytic hierarchy process,”
Mathematical Modelling, p. 16, 2008.
[17] S. Kirby and N. Pilon, “Minimizing the Cost of Delay for Airspace
Users,” in 12th USA/Europe Air Traffic Management Research and
Development Seminar (ATM2017), Seattle, 2017.
[18] University of Westminster, “Thematic Challenge 4 - EUROCONTROL
Engage KTN,” Eurocontrol, Tech. Rep., 2019.
